2
\$\begingroup\$

Recently noticed my GPU card (Nvidia 1060) made more coil noise (electrical noise, noise level proportional to frame rate) than usual. I discovered that directly behind the GPU chip one of 2 big 470uF, 2 volt capacitors had broken off. I just received 3 replacements (EEF-HX0E471R4).

But it got me thinking about the noise I heard and the remaining capacitor. Should I change the other capacitor also?

The voltage ripple must be bigger with the loss of capacitance, but can the system be designed such that a loss of capacitance can start a domino like degradation of the rest of the capacitors? (I am an EE, and of course I have to overthink it, but I have little experience with ~2 GHz digital design)

enter image description here

\$\endgroup\$
4
  • 2
    \$\begingroup\$ In my experience, these kind of cards don't need a lot of excuses to activate screaming coils. Without knowing what the caps are for, answers will just be speculation. 470uF doesn't sound like anything related to RF, but more likely to voltage regulators, bulk caps etc. It might be more meaningful to examine why your caps decided to leave the PCB. \$\endgroup\$
    – Lundin
    Nov 3 '20 at 11:27
  • \$\begingroup\$ The others may have been damaged by what damaged the first one, such as an impact. But with no information to hand and in the dark, consider the consequences of replacing vs not replacing, assuming you can solder competently. Replacing when it doesn't need it wastes the capacitors cost (£1.44), which you've already spent on these extra caps anyway. Not replacing them when they do need it could possibly lead to damage of other parts, which cost a lot more than £1.44. If uncertain, replace the second one with the parts you've already bought anyway. \$\endgroup\$
    – TonyM
    Nov 3 '20 at 11:31
  • \$\begingroup\$ Well since the capacitor is directly under the GPU and it is a very low ESL/ESR type I still expect it to have an impact. But it seems more proportional with frame rates (60-144Hz) than the actual GPU clock rates. It is bursts of 2 GHz @ 100Hz, but I am not sure that makes sense in terms of sound output. \$\endgroup\$
    – Bonnevie
    Nov 3 '20 at 11:38
  • \$\begingroup\$ @Bonnevie I would imagine that it is related to some internal voltage regulator inside the GPU. As for what causes screaming coils, it could be some particular feature getting activated. \$\endgroup\$
    – Lundin
    Nov 3 '20 at 11:46
3
\$\begingroup\$

If one of the bulk capacitors was missing it will increase the ripple current the other capacitors are seeing.

This increase might exceed the rated value, which puts it into the realm of undefined behavior (sudden unexpected things might happen).

But increase of ripple current always leads to increased self heating which decreases the expected life of the capacitor.

So I'd replace them all while you are at it - just to be on the safe side (you can't fix the graphics processor if that dies next time).


This document from Nippon Chemicon shows a nice overview of possible failure modes with causes in aluminium electrolytic capacitors: Failure modes

So the main fault mode is the one I described before, but there is a dotted line which indicates that it might be possible for excessive ripple current to lead to a short circuit condition, which would probably end quite bad for the card.


And the datasheet of the replacement part has this note in it:

Ripple current note of the capacitor

\$\endgroup\$
2
  • \$\begingroup\$ Alu electrolytes for PSU filters readily displays maximum AC ripple in their specs due to self heating, but is that a concern in this kind of decoupling? I would never consider self-heating in MLCC for example. \$\endgroup\$
    – Bonnevie
    Nov 3 '20 at 13:16
  • \$\begingroup\$ @Bonnevie I am not 100 % sure on that, but the ESR will not go down to 0 even if you increase the frequency to 2 GHz, so it will still heat up. I have no idea how large the ripple currents are that those caps see - the hordes of MLCC around should take most of it, but at the current levels of a GPU there might still be a lot left for the bulk caps. \$\endgroup\$
    – Arsenal
    Nov 3 '20 at 13:38
2
\$\begingroup\$

"Capacitor broken off" sounds more like a mechanical failure than an electrical one; either it was badly soldered, or damaged during shipping and installation, or it's been jammed against one of the cable bundles inside the PC and levered off.

(A photo might help us see this)

It's likely to be a bulk decoupling capacitor. Losing one of these increases the ripple in the power supply, and therefore very slightly increases the throughput of coulombs in and out of the other bulk capacitors as they try to compensate. I wouldn't expect this to cause them to fail, but it increases the probability of the card glitching under heavy load or dips in the supply from the PSU.

There was, in the early 2000s, a "capacitor plague" as a result of manufacturing defects, but the symptom there was bursting and leaks.

\$\endgroup\$
2
  • \$\begingroup\$ I am pretty sure it fell off because I rough handled it. It has no backplate and I just put it on a hard wood table expecting the heat sink screws to be the feet. \$\endgroup\$
    – Bonnevie
    Nov 3 '20 at 11:35
  • \$\begingroup\$ I added a picture. It is right behind the GPU so it must be important. \$\endgroup\$
    – Bonnevie
    Nov 3 '20 at 13:20

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.